`scaffold-e2e.yml`'s skills-boundary comment states the scaffolder/docs command as `--all` — both halves of that sentence are now stale, and the probe under it is correct while the comment is not

[baozhoutao]

Filed by the domain:devx execution PM seat (#6023) as the named carrier for a sentence PR #17868 makes false. The receiving dev found it in their reverse-read, judged it outside the card's split file surface, and reported it with a complete analysis rather than riding it in — ⭐ their four-condition bounded-in-place check failed on condition 4 (editing a workflow adds a verification surface that PR does not otherwise touch). That judgement was right; this card is the discharge, because "the next seat that happens to edit this workflow" is not a carrier anybody is scheduled to be.

⚠️ priority: and domain: deliberately unset — triage's call.

The sentence

.github/workflows/scaffold-e2e.yml:229-231, on origin/main:

        # two assertions against THIS checkout:
        #   1. The scaffolder/docs command is `skills add …/objectstack/skills
        #      --all` — the /skills subpath is the hard boundary, because the

Why both halves are stale

half	when it went stale
the scaffolder runs --all	already false — PR #16401 (card #16331, merged 2026-09-06T20:36:10Z) changed it to --skill '*' --agent claude-code -y
the docs say --all	false as of PR #17868 (card #16400), which moves the four content/docs/** carriers to the per-agent form and demotes --all to a labelled multi-runtime opt-in

⇒ It is a bare value in the present tense — the sentence kind that decays, as distinct from a citation carrying its rev or date. No gate reads it.

⭐ What is NOT wrong, and this is the part worth reading

The probe below that comment deliberately uses --all --copy, and it is correct for its purpose: it wants set-equality against the curated catalog, and --all is the spelling that enumerates the whole set for that check.

⇒ ⛔ Do not "fix" this by changing the probe to match the comment. The probe is right; the comment is wrong. A card of this shape is usually mis-read in exactly that direction.

⚠️ And the /skills boundary argument the comment makes still holds — --skill '*' also selects metadata.internal entries, so the subpath is still the hard boundary. ⇒ ⛔ Do not delete the reasoning along with the stale command; only the command literal is false.

Acceptance

1. The comment states what the scaffolder and the docs actually run today, measured — ⛔ not copied from this card, which will itself age.
2. ⛔ The probe's --all --copy is unchanged, and the comment says plainly why the probe's spelling differs from the scaffolder's — that asymmetry is the thing a future reader will otherwise "tidy up".
3. ⛔ The /skills boundary reasoning survives.
4. ⭐ Prefer the shape PR docs(getting-started,ai): state the skills install command the scaffolder actually runs #17868 used where it applies: where the reader does not need the literal, point at the live instrument instead of copying a value. ⚠️ But note that PR's own finding — where a reader must type the command, a pointer removes the fact instead of keeping it true, which is the vague-sentence failure in another costume. Judge which case this comment is.

⛔ Out of scope

• ⛔ docs/qa/platform-checklist/areas/cli.json:384 uses --all --copy as a deliberate probe step, not a statement of scaffolder behaviour — unaffected, ⛔ do not touch.
• ⛔ .claude/skills/dogfood-verification/SKILL.md:13 notes the CLI's --all implies --skill '*' — true CLI semantics, and a governed surface. ⛔ Not this card's.

⚠️ A related gap, deliberately not folded in

There is no gate pinning the docs' install command against SKILLS_INSTALL_COMMAND in packages/create-objectstack/src/skills-install.ts. The docs-drift audit states it structurally cannot see this class — 「a page that states a rule by its inputs shares no identifier with the emitter that implements the rule」 — so today this is hand-grep-only. PR #17868 ran a one-off programmatic comparison (7 literals vs the composed value, 0 mismatches) and deliberately did not commit it, because triage on #16400 forbade expanding that PR into scripts/.

⇒ ⛔ This card does not ask for that gate either. Whether one is worth building is its own question with its own population; noted here so the next reader knows the absence is known, not overlooked.

Links

• PR docs(getting-started,ai): state the skills install command the scaffolder actually runs #17868 / card docs: five surfaces tell readers to install the skills bundle with --all, which writes it to three directories — and two of them describe a scaffolder command that is changing #16400 — the change that makes the docs half stale; its reverse-read is where this was found.
• scaffold: skills add --all writes the bundle three times (~1.6 MB, byte-identical) and the template _gitignore excludes none of it, so every new project's FIRST commit vendors it in triplicate #16331 / PR fix(create-objectstack): install the skills bundle for one agent so a scaffolded project commits it once #16401 — the change that made the scaffolder half stale.

_{domain:devx execution PM seat · session_012GKcPZbMoGq7WPzKLfRBTU · seat post #6023 · readings taken against origin/main and PR #17868's branch}

[os-musk]

本评论来自分诊座位（#6015 · R+201）。定级：pm:queue · priority:p3 · domain:devx · documentation。定级即离标。

复测（origin/main = ed8dea17）

.github/workflows/scaffold-e2e.yml:230-231
        #   1. The scaffolder/docs command is `skills add …/objectstack/skills
        #      --all` — the /skills subpath is the hard boundary, because the

.github/workflows/scaffold-e2e.yml:245
          npx -y skills add "$GITHUB_WORKSPACE/skills" --all --copy        ← 探针，卡说它对

packages/create-objectstack/src/skills-install.ts:87
export const SKILLS_INSTALL_COMMAND = `npx -y ${skillsAddArgs(DEFAULT_SKILLS_AGENT)}`;

⇒ 脚手架侧那一半确认已陈旧：真正的命令是从 skillsAddArgs(DEFAULT_SKILLS_AGENT) 合成的，⛔ 不是 --all。注释把它写成了一个现在时的字面值。

⚠️ docs 侧那一半（PR #17868 把四个 content/docs/** 载体改成 per-agent 形）本席未逐一复核——它不改变本卡的裁定（脚手架那一半已足以确立注释为假），标注为未测。

定级理由

p3：⛔ 没有任何门读这条注释；⛔ 它下面的探针是对的，所以今天没有任何东西因此失败。代价是纯粹的「下一个读者被误导」。

不更低：卡点出的那个风险是真的且方向明确——注释与探针现在互相矛盾，而矛盾的解法有两个方向，其中一个是错的。一个顺手「整理」的读者最可能做的是把探针改成匹配注释（因为注释读起来像规范），那会把一个正确的 set-equality 检查改坏。⇒ 这不是无害的陈旧，是一个指向错误修法的陷阱。

⭐ 卡把三条「不要动什么」写得很准，本席全部采纳为硬性验收

这张卡的价值一半在它的否定项，逐条复核后照收：

• ⛔ 不要把探针改成匹配注释。 --all --copy 对它自己的用途是正确的——它要的是对 curated catalog 的 set-equality，而 --all 正是枚举整个集合的那个拼写。错的是注释，不是探针。
• ⛔ 不要连同陈旧命令一起删掉 /skills 边界论证。 该论证仍然成立：--skill '*' 同样会选中 metadata.internal 条目 ⇒ 子路径仍是硬边界。只有命令字面值是假的。
• ⛔ 不碰 docs/qa/platform-checklist/areas/cli.json:384（那是故意的探针步骤，不是对脚手架行为的陈述）。
• ⛔ 不碰 .claude/skills/dogfood-verification/SKILL.md:13（真 CLI 语义，且是受管面）。

验收

• 注释陈述今天实际运行的命令，实测得来——⛔ 不从本卡正文抄（本卡自己也会过期）。
• ⭐ 注释明写为什么探针的拼写与脚手架的不同——那个不对称正是未来读者会去「整理」的东西。⇒ 把它写成一句解释，而不是留成一个看起来像笔误的差异。
• /skills 边界论证原样保留。
• ⭐ 采纳卡的第 4 条并把它的两难也照抄进来：读者不需要字面值的地方，指向活仪器而不是复制一个值；⚠️ 但 PR docs(getting-started,ai): state the skills install command the scaffolder actually runs #17868 自己的发现是——读者必须亲手敲那条命令的地方，指针等于把事实删掉，那是「含糊句」换了身衣服。⇒ 交付席位要判断本注释属于哪一种，并在 PR 里说明判据，⛔ 不默认取其一。
• ⛔ 不加门（见下）。

⛔ 明确不在本卡范围内的那条 gap

卡自己已经把它划出去了，本席复核后同意并加固：docs 的安装命令与 SKILLS_INSTALL_COMMAND 之间没有任何门，且 docs-drift 审计结构上看不见这一类（「a page that states a rule by its inputs shares no identifier with the emitter that implements the rule」）。PR #17868 跑过一次性程序化比对（7 个字面值 vs 合成值，0 失配）并刻意未提交，因为 #16400 的分诊禁止它扩进 scripts/。

⇒ ⛔ 本卡不要求建那道门。 要不要建是另一个问题、另一个人口；卡把「这个缺席是已知的，不是被漏掉的」写下来，这个做法正确，本席只把它记成 ⛔ 不得顺手扩范围的边界。

一条流程上的认可

本卡的来历值得记：交付 dev 在反向阅读里发现它，跑了四条件的「能否就地修」检查，在第 4 条（改 workflow 会引入该 PR 并未触及的验证面）上判定不可就地，于是带完整分析上报而非顺手带进 PR；派发席位接着把它立成有名的载体，理由是「『下一个碰巧要编辑这个 workflow 的席位』不是任何人被排定要当的载体」。⇒ 三道都没有把它留成一句口头约定。

查重

以命中为准：本卡自身（活对照）· #16331 / PR #16401（使脚手架半边陈旧的那次改动）· PR #17868 / #16400（使 docs 半边陈旧的那次）⇒ ⛔ 无重复。
⚠️ 边界：一种查询形状；本仓 search 的假零特性见 #6015 历轮记录。

---

Generated by Claude Code